library("lvnet")
library("psych")
library("lavaan")
data(HolzingerSwineford1939)
Data <- HolzingerSwineford1939[,7:15]
Lambda <- matrix(0, 9, 3)
Lambda[1:3,1] <- NA
Lambda[4:6,2] <- NA
Lambda[7:9,3] <- NA
CFA <- lvnet(Data, lambda = Lambda)
Error in if (any(select)) { : missing value where TRUE/FALSE needed

Was simply trying to work through the example on the readme but had the following error pop up.

Hi Sacha,

(I subscribed to GitHub!)

Question: How does lvnet determine the degrees of freedom of a model?

The question follows from the following: Below I fit a confirmatory network model on raw data. The model is saturated, so I expect the chi-square to be 0 with 0 degrees of freedom. According to the outcome of lvnet, the chi-square value is 0 indeed, but it reports a df of 9. That is understandable in the light of the number of parameters lvnet reports, namely 1, but I disagree with that number: lvnet estimated 4 omega_deltas and 6 omega_thetas, so parameters 10 in total (although the parameterization is in theta_inverse, if I understand correctly ). This would make the dfs equal 0.

setup

no <- 250
nv <- 4

mvec <- rep( 0,nv )

cmat <- matrix( scan(), nv, nv, TRUE )
4.3685800 3.9192711 0.8919471 2.427868
3.9192711 6.0460335 0.9923974 2.517379
0.8919471 0.9923974 2.1397640 1.317591
2.4278675 2.5173785 1.3175909 35.906516

simulate

dat <- MASS::mvrnorm( no, mvec, cmat,emp = TRUE )

fit (saturated) network model

check <- lvnet(dat, omega_theta = diag( 0, nv ) / diag( nv ) )

result

( estpcor <- check$matrices$omega_theta + diag( nv) )

compare with:

cor2pcor( cmat ) # OK!

chi-square (should be 0)

round ( check$fitMeasures$chisq, 8 ) # OK!

dfs (should be 0)

check$fitMeasures$df # Err.... not OK!